Kidney Week

Abstract: FR-PO0004

Machine Learning as a Predictor of Pediatric Renal Osteodystrophy Using Biomarkers and Anthropometric Parameters Reflects Influence of Iron Status and Inflammation on Bone Mineralization and Turnover

Session Information

• Artificial Intelligence and Digital Health at the Bedside
  November 07, 2025 | Location: Exhibit Hall, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Artificial Intelligence, Digital Health, and Data Science

• 300 Artificial Intelligence, Digital Health, and Data Science

Authors

• Tiu, Ryan Alyson-Yao, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Ryan Alyson-Yao Tiu

• Feng, Jeffrey, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Jeffrey Feng

• Ganz, Tomas, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Tomas Ganz, MD, PhD

• Pereira, Renata C., University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Renata C. Pereira, PhD

• Gales, Barbara, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Barbara Gales, RN

• Bui, Alex, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Alex Bui, PhD

• Salusky, Isidro B., University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Isidro B. Salusky, MD, FASN

Background

The complex crosstalk between iron dysregulation, inflammation, chronic kidney disease (CKD), and bone health is still being elucidated. Using biomarkers relevant to these etiologies, we developed a machine learning (ML) model to predict bone turnover (BT) and mineralization (BM) in pediatric CKD-mineral and bone disorder.

Methods

This cross-sectional study includes 659 iliac crest bone biopsies performed in pediatric patients from 1983-2023 (Fig. 1a). None of the patients were treated with aluminum binders. PTH (IRMA 1st & 2nd generation, Immutopics^R, San Clemente, CA), comprehensive metabolic panel, blood count, iron studies, height, and weight were measured at biopsy. All biopsies were read by the same pathologist (RP) using OsteoMetrics. ML models were used to predict BT (defined by bone formation rate) and BM (defined by osteoid thickness and mineralization lag time (MLT)) from biomarkers, height, and weight. SHapley Additive exPlanations (SHAP) facilitated feature importance analysis.

Results

On a holdout test set, random forest achieved the highest area under the receiver operating curve of 0.69 (95% CI 0.59-0.78) and 0.83 (95% CI 0.74-0.90) for predicting BM and BT, respectively. Among the top features predicting abnormal BM were low height, high hepcidin, and low iron saturation. When available, hepcidin was often the top predictive feature, with high values predicting abnormal BM (Fig. 1b, c). Hepcidin correlated with MLT (r=.3, P=.01). Top features predicting high BT were high PTH, low iron stores, and high creatinine, which was correlated with osteoclast surface/bone surface (r=.5, P<.001), a histomorphometric indicator of BT.

Conclusion

ML shows potential in predicting BT and BM and highlights the interplay of declining GFR, the inflammatory milieu of CKD, and iron dysregulation on bone and mineral metabolism.

Digital Object Identifier (DOI)

• doi: 10.1681/ASN.2025f7dcz1eg